Dynamic, Individualized, Artificial Intelligence-Powered Prediction of Recurrent Autoimmune Hepatitis after Liver Transplantation: A Multi-Centre Cohort Study

Project Overview

This project implements an end-to-end machine learning workflow that processes data, trains predictive models, and provides detailed model interpretability through SHAP analysis. Our AI-powered clinical decision model provides personalized prediction of recurrent Autoimmune Hepatitis(rAIH) post liver transplant. It provides insight into modifiable and non-modifiable predictors of rAIH, supporting more personalized immunosuppressive strategies to improve long-term outcomes.

Repository Structure

.
├── requirements.txt      # All packages required for the Python environment
├── data_processing.py    # Data loading, cleaning, and preprocessing
├── models.py             # Machine learning model definitions and training
├── shap_analysis.py      # SHAP-based model interpretability analysis
├── models                # Trained models 
└── README.md             # Project documentation

File Descriptions

data_processing.py

Handles all data-related operations including:

• Data loading from various sources
• Data cleaning and validation
• Feature engineering
• Train-test splitting
• Data transformation and scaling

models.py

Contains machine learning model implementations:

• Model architecture definitions
• Training procedures
• Hyperparameter configuration
• Model evaluation metrics
• Model saving and loading utilities

shap_analysis.py

Provides model interpretability tools:

• SHAP value computation
• Feature importance visualization
• Individual prediction explanations
• Summary plots and dependence plots
• Model behavior analysis

SHAP Analysis Outputs

The SHAP analysis module generates several types of visualizations:

• Summary Plots: Overview of feature importance across all predictions
• Dependence Plots: Relationship between feature values and SHAP values
• Force Plots: Individual prediction explanations

Example Results

After running the complete pipeline, you can expect:

• Trained model with performance metrics
• SHAP values for all predictions
• Visualization plots saved to the output directory
• Feature importance rankings

Configuration

Adjust parameters in each module:

• Data Processing: Modify feature engineering steps, scaling methods
• Models: Change model types, hyperparameters, cross-validation settings
• SHAP Analysis: Adjust plot styles, sample sizes, visualization options